/
validate.py
165 lines (123 loc) · 4.96 KB
/
validate.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
import os
import re
import sys
import time
import json
import pdb
import numpy as np
import tensorflow as tf
# [TODO]
import matplotlib as mpl
mpl.use('Agg')
import matplotlib.pyplot as plt
from scipy.io import loadmat
from datetime import datetime
from vae.model import VAE2
from iohandler.spectral_reader import vc2016TFWholeReader
from util.spectral_processing import MelCepstralProcessing
FLAGS = tf.app.flags.FLAGS
# tf.app.flags.DEFINE_string('source', 'SF1', 'list of source speakers')
# tf.app.flags.DEFINE_string('target', 'TM3', 'list of target speakers')
# tf.app.flags.DEFINE_string('{:s}-{:s}-trn', '', 'data dir')
tf.app.flags.DEFINE_string(
'datadir', '/home/jrm/proj/vc2016b/Parallel_Frames', 'data dir')
tf.app.flags.DEFINE_string(
'architecture', 'architecture.json', 'network architecture')
tf.app.flags.DEFINE_string('logdir', 'logdir', 'log dir')
# tf.app.flags.DEFINE_string(
# 'logdir_root', None, 'log dir')
# tf.app.flags.DEFINE_string(
# 'restore_from', None, 'resotre form dir')
tf.app.flags.DEFINE_string('checkpoint', None, 'model checkpoint')
tf.app.flags.DEFINE_integer('batch_size', 128, 'batch size')
tf.app.flags.DEFINE_float('l2_regularization', 0.0, 'L2 regularization')
tf.app.flags.DEFINE_float('lr', 1e-3, 'learning rate')
tf.app.flags.DEFINE_integer('num_steps', 10000, 'num of steps (frames)')
tf.app.flags.DEFINE_integer('source_id', 0, 'target id (SF1 = 1, TM3 = 9)')
tf.app.flags.DEFINE_integer('target_id', 9, 'target id (SF1 = 1, TM3 = 9)')
tf.app.flags.DEFINE_string(
'file_filter', '.*\.bin', 'filename filter')
# TEST_PATTERN = 'SF1-100001.bin'
TEST_PATTERN = '.*001.bin'
N_SPEAKER = 10
mFea = 513 + 1
# ['SF1', 'SF2', 'SF3', 'SM1', 'SM2', 'TF1', 'TF2', 'TM1', 'TM2', 'TM3']
def main():
if FLAGS.checkpoint is None:
raise
# FLAGS
started_datestring = "{0:%Y-%m%d-%H%M-%S}".format(datetime.now())
logdir = os.path.join(FLAGS.logdir, 'generate', started_datestring)
# with open(FLAGS.)
with open(FLAGS.architecture) as f:
architecture = json.load(f)
# coord = tf.train.Coordinator()
print(FLAGS.datadir)
# label, spectrum, filename = vc2016TFWholeReader(
# datadir=FLAGS.datadir,
# pattern=TEST_PATTERN,
# output_filename=True)
net = VAE2(
batch_size=128, # [TODO] useless attribute?
architecture=architecture)
x_ = tf.placeholder(tf.float32)
y_ = tf.placeholder(tf.float32)
z_ = net.encode(x_)
xh_ = net.decode(z_, y_)
# Restore model
sess = tf.Session()
variables_to_restore = {
var.name[:-2]: var for var in tf.all_variables()
if not ('state_buffer' in var.name or 'pointer' in var.name)}
saver = tf.train.Saver(variables_to_restore)
print('Restoring model from {}'.format(FLAGS.checkpoint))
saver.restore(sess, FLAGS.checkpoint)
# threads = tf.train.start_queue_runners(sess=sess, coord=coord)
evaluator = MelCepstralProcessing(
itype=3, order=24, isMatrix=True, drop_zeroth=True)
# file_indices = range(150, 163)
speakers = ['SF1', 'SF2', 'SF3', 'SM1', 'SM2', 'TF1', 'TF2', 'TM1', 'TM2', 'TM3']
# src_spk = 'SF1'
# trg_spk = 'TM3'
src_id = FLAGS.source_id
trg_id = FLAGS.target_id
src_spk = speakers[src_id]
trg_spk = speakers[trg_id]
test_sentences = range(150, 163)
mcds = np.zeros((len(test_sentences),))
num_frame = np.zeros((len(test_sentences),))
for i in test_sentences:
src_file = os.path.join(
FLAGS.datadir,'{}/{:6d}.mat'.format(src_spk, 100000 + i))
trg_file = os.path.join(
FLAGS.datadir,'{}/{}/{:6d}.mat'.format(src_spk, trg_spk, 100000 + i))
# # *.bin reader
# x = np.fromfile(src_file, dtype=np.float32)
# x = np.reshape(x, [-1, mFea])
x = loadmat(src_file)['sp'].T
y = np.zeros([x.shape[0], 10])
y[:, trg_id] = 1.0
xh = sess.run(xh_, feed_dict={x_: x, y_: y})
xh = np.power(10, xh)
xh = xh / xh.sum(1).reshape([-1, 1])
xh = xh.astype(np.float64)
# x = np.power(10, x)
# x = x.astype(np.float64)
x_src = np.power(10, x)
x_src = x_src.astype(np.float64)
x_trg = loadmat(trg_file)['sp'].T
x_trg = np.power(10, x_trg)
x_trg = x_trg.astype(np.float64)
mcd_after = evaluator.mmcd(xh, x_trg)
mcd_before = evaluator.mmcd(x_src, x_trg)
print('{:.2f} -> {:.2f}'.format(mcd_before, mcd_after))
mcds[i - test_sentences[0]] = mcd_after
num_frame[i - test_sentences[0]] = x_src.shape[0]
# mcd.append([mcd_after, x_src.shape[0]])
# Note:
# 1. Note it's "log10" (so how to compute MCD?)
mcds = num_frame * mcds
mcds = mcds.sum() / num_frame.sum()
print('Mean MCD: {:.2f}'.format(mcds))
if __name__ == '__main__':
main()